P4B533 & Latest 1015 BIOS => could it now be "800mhz" ready ? - ABXZone Computer Forums
Hi there,
Have been playing with my P4B533 and a new 800MHz P4 2.8C the last two days. As others in this forum already mentioned, the computer posts and boots into Windows XP without any problem. You'll have to set the external clock in the BIOS manually to 200MHz, 1:1 memory ratio (4:3 doesn't seem to have any effect), check the Hyper-threading setting (ON by default) and select the most relaxed memory timings (by SPD or 2.5/3/3/7 for my 512MB PC3200 Crucial stick - otherwise it won't post).
If Hyper-threading is turned on, XP will find a new ACPI device (the "logical" 2nd processor), asks to restart and voila - everything SEEMS to be working fine.
Computer runs stable, Device Manager has two processor entries and the Windows Task Manager shows two separate CPU usage windows. CPU-Z confirms that the processor runs at 2.8GHz, and the memory at 400MHz with 2.5/3/3/7 timing.
SANDRA MAX3! runs all the benchmarks, here are some samples (w/ Hyper-threading turned on and all module options checked):
*CPU Arithmetic: Dhrystone ALU 8440 MIPS, Whetstone FPU/iSSE2 2420/5400 MFLOPS
*CPU Multi-Media: Integer iSSE2 12927 it/s, Float iSSE2 20354 it/s
*Memory Bandwidth: Int Buff iSSE2 2925 MB/s, Float iSSE2 2926 MB/s
*Cache & Memory (fastest/slowest): 4kB Blocks 30605 MB/s, 256MB Blocks 1500 MB/s
I also ran all tests in Memtest86 an entire night (~8hrs) without finding any errors. I don't have a decent 3-D card in this computer yet, so I can't comment on any stability issues with 3DMark or heavy gaming yet (it is my girlfriend's, but I will move my SCSI gear and GeForce card over soon and give her my old rig instead).
Well, here is the problem (as you might have guessed): The PRIME95 torture test and the 32M digit calculation with Super PI exhibit rounding errors within the first hour.
This is both with hyper-threading disabled and only one instance of PRIME95, and hyper-threading enabled and two instances of PRIME95 (I think Super PI only allows one instance at a time). With the CPU load at 100%, ASUS PC Probe shows that the CPU temperature rises to around 51°C, 10-12°C above the MB temperature and 25°C above ambient. Considering that the cooling inside my In-Win S508 case sucks (I have only one extra fan in the rear), I guess that these temperatures look pretty normal (both the CPU and MB are around 10°C above ambient when idle).
First thing I did was to replace my 300W Sparkle power supply with a 430W Antec TruePower - didn't make any difference (the rails read the same as before idle and under load). I briefly tried upping VCore and/or the DDR voltage from their defaults (1.525V/2.6V), but this didn't help, either. This is not really a surprise, since I am neither overclocking the CPU nor the RAM. At this point I was fairly confident that the north bridge is causing the problems since it is running fairly outside of its specs (at 200MHz it is technically 50% overclocked). The passive heat spreader got only mildly warm to the touch, so I unclipped it from the board, removed the pink goo and the foam pad around, applied a thin layer of silver grease and put the sink back on: Again, PRIME95 and Super PI detected hardware failures (rounding errors) after a while. Not sure if I was dreaming, but I think that the MB temperature dropped a degree or three (where is the sensor located, anyway?)
Next, I started reducing the FSB: First 199MHz - Nope, then 198MHz, Bingo!
Two instances of PRIME95 ran for a couple hours without complaining. Super PI finished the 32M calculation. Damn, this was so close to the processors rated speed, I had to find a way around the memory controller limitation. Our forum moderator bigtoe posted a review about the P4G8X back in January, and he suggested that upping the AGP voltage on that board will also increase the voltage to the chipset which in turn allows for higher clock-speeds. I guess the same "trick" works for the P4B533! 
Increased AGP voltage from 1.5V to 1.6V (all other settings in the BIOS relating to the processor and chipset on Default or AUTO) and everything is fine. Super PI takes exactly 48 minutes, 35 seconds to do the 32M digit calculation (512MB RAM, Windows paging file disabled) and two instances of PRIME95 ran happily for 8+ hours. Your mileage may vary, of course.
Hope this helps,
Buhntz
Current specs of this test system:
P4B533 (845E chipset), PCB Rev. 1.03, BIOS version 1015
P4 2.8C (800MHz)
512MB Crucial PC3200 (CL=3 • Unbuffered • Non-parity • 5ns • 2.5V • 64Meg x 64)
Antec TruePower 430W
3dfx Voodoo 3 2000 PCI (don't laugh - it's my girlfriend's)
Western Digital 60GB Ultra ATA/100
Plextor PX-32TS SCSI CD-R
ASUS SC-200 SCSI controller (doesn't allow booting from CD with this MB)
In-Win S508 case
Windows XP Pro w/ SP1
沒有留言:
張貼留言